Statement of Dr. Gilbert Weigand,
Deputy Assistant Secretary for Research, Development and Simulation
Office of Defense Programs
U.S. Department of Energy
Mr. Chairman and distinguished members of the Committee, thank you for inviting me to testify on security issues. In my capacity as Deputy Assistant Secretary for Research, Development and Simulation for the Office of Defense Programs, I oversee nuclear weapons research, development and simulation work at Los Alamos, Lawrence Livermore and Sandia National Laboratories and the Nevada Test Site. While my office had no direct role in the sale and recovery of the Sandia Intel Paragon computer, I will try to provide you with information from the Department about that incident. I am best prepared to address the capabilities of the computer itself, and computers in general, with regard to their use in the U.S. nuclear weapons program, although to answer your questions on computers in some degree of detail, it may be necessary to speak in a classified setting.
The Department named an Incident Review Team to conduct an investigation of the sale and recovery of the Intel Paragon computer. Experts on procurement and export controls from the Departments Office of Contract and Resource Management and the Office of Arms Control and Nonproliferation, with background in procedures and requirements regarding such transactions were asked to perform the investigation and to provide the Department with a report of their findings. I would ask that a copy of that report, dated September 23, 1999, titled Sandia National Laboratories Sale of an Intel Paragon Supercomputer be entered into the hearing record as part of my testimony today.
The report provides a step by step analysis of how the disposal of the Intel Paragon was handled, according to established procedures, but was still not subjected to the kind of analysis that might have stopped the sale before it was completed. Among the findings was that a thorough export control review was not performed prior to offering the machine for sale to the public. Also, the purchaser, Mr. Jiang, himself identified several manuals, disks, and Sandia binders all related to the Paragon, that should not have been part of the sale, and when the government repurchased the computer, revealed that a complete inspection of the computer packing boxes at the time of inventory had not been performed. The report also makes a number of recommendations and comments for the future with regard to: communications, process; policy and training of relevant staff; and security sensitivities.
I fully supported the Departments decision to investigate the Sandia Intel Paragon incident and my office provided support to the experts who were assigned to the team. DOEs Office of Arms Control and Nonproliferation led recovery activities and was supported by the Office of Contract and Resource Management. Experts in the Office of Arms Control and Nonproliferation, as well as field and laboratory staff, provided the input for the incident investigation report.
The Intel Paragon was a five year old computer that had been declared obsolete by Sandia National Laboratories (SNL) in April of 1998. After determining that there were no U.S. government agencies interested in the system, it was sold by Sandia on September 29, 1998 to a licensed American corporation, EHI Group USA, located in Cupertino, California. Months after the sale and transfer of the computer, it came to the attention of Sandia that a principal in the company, Mr. Quingchang Jiang, was a Chinese national. Sandia had acquired the machine for $9 Million in the early 1990's. The sale price to EHI was $30,088. At the time of sale, the computer was disassembled into 21 components. Intel reported that Mr. Jiang tried to obtain additional parts for the machine on several occasions, but Intel did not sell him any parts. The machine was retrieved July 21, 1999 and it was transported to Livermore, California and placed in a secure storage warehouse physically located on the property of Sandia National Laboratories in Livermore, California, where it remains today, in a secure area with limited access.
Depending on its operating configuration, the Intel Paragon was capable of up to 150,000 million theoretical operations per second (MTOPS) and 200,000 MTOPS. While the theoretical speed sounds impressive, in practical terms, or in terms of applications, it was not as impressive because of the limitations that exist within the machine. It was a forerunner research machine used, in part, to learn how to build next generation machines. Today, it would be most useful in an academic setting for research on parallel computer machines. Another use for it would be for spare parts to use on a similar machine.
This machine would be very expensive to operate and not very efficient were anyone to try to use it in what we refer to as "production mode" which basically means in regular operation, the way we use computers in an office. It did have some of the U.S.s most leading edge interconnect technology and was sold with its operating system still on the machine, although the disk drives on which classified information had been processed were not sold with the machine. To a nation that had little computer power or capability, reassembling this machine could have provided significant capability.
Paragon Usage at Sandia
The Intel Paragon was an early generation of massively parallel computers and one of the first in the world with as many as 1872 computing nodes. It was primarily used to advance the state-of-the-art in computing science and computational science associated with scaling system software and numerical algorithms to such a large number of nodes. Funds to purchase the system were partially provided by DOE/Defense Programs, partially by the Defense Advanced Research Projects Agency (DARPA), and partially by the U.S. Air Force, all as part of the National Consortium for High Performance Computing. Application by Sandia and other consortium members dealt primarily with computational science issues related to defense applications, including:
Computer science research conducted by Sandia and other consortium members included:
The history of the Intel Paragon is that in 1993, Sandia took delivery of a 140 GFlops (billion floating point operations per second) peak performance Paragon supercomputer, Model XPS 140, manufactured by Intel Corporation. The supercomputer consisted of 1344 nodes each with .016 Gbytes (gigabytes) of memory, plus 528 nodes each with .032 Gbytes of memory per nodes, for a total of 1872 nodes and an aggregate 38.4 Gbytes main memory. Each node contained two Intel i860 microprocessors with a peak performance of .075 GFlops each. In general operation one processor was used for computation and the other processor was dedicated to inter-node communication. The aggregate peak performance in this mode was 140.4 GFlops. All nodes were inter-connected in a two-dimensional mesh topology consisting of 117 rows each with 16 nodes (16 x 117 mesh). Each mesh link was capable of transferring data at 0.2 Gbytes per second in each direction simultaneously. The inter-node message latency (due to both software and hardware) from memory on one node to memory on another node was approximately 60 micro-seconds. Approximately 200 Gbytes of disk storage were embedded in the system.
In summary, the Sandia Intel Paragon sale appears to have resulted from a combination of lack of communication combined with some unfortunate assumptions. We successfully retrieved the machine before it could be reassembled and operated or shipped abroad. Indeed, there is no indication that there was ever any attempt to reassemble or export the machine. We have investigated what happened, and how, and have developed a solid path forward, as I think the report makes clear.